Um estudo sobre os efeitos da utilização do ECN no desempenho das redes Diffserv

Abstract

This dissertation presents a study that evaluates the performance of an architecture that aggregates the Differentiated Service (DiffServ) model and the Explicit Congestion Notifícation (ECN) mechanism. These technologies represent two key elements for Quality of Service (QoS), which are: Service differentiation and traffic management. A series of simulations were implemented, not only for the proposed model, but also for the traditional one, and the obtained results were compared in order to identify the solution that offers better performance and QoS. The parameters analyzed were link effíciency, fairness of bandwidth sharing and packet discard, timeout and retransmission rates. Issues that affect the Service levei have also been assessed like: number of active flows, link congestion levei, interaction with non responsive sources, flows with distinct RTTs, variation in packet size, number of flows in an aggregate, size of target rate and interaction with varied congestion control algorithms. The simulations results indicate that the use of ECN optimizes DiffServ performance for a great variety of scenarios. In particular, when the network is correctly dimensioned, the use of ECN, not only provides performance benefits, but also optimizes bandwidth sharing and minimizes the effects of most factors that affect the Service levei provided to sources with distinct characteristics. This behavior is even more evident when the total reserved bandwidth approaches the total link capacity, but it is reversed when the network is underprovisioned. This demonstrates that, to make full usage of the benefits of the DiffServ and ECN combination, it is essential to avoid that the total contracted bandwidth be superior to the link capacity.